Open-weight pressure mounts on closed models

The economics of closed models are bending under open-weight pressure. OpenAI's public concern about Chinese open-weight models isn't geopolitical posturing—it's unit economics. When open models can be served at $0.12 per million tokens versus closed providers at $2.50, the arbitrage becomes structural. Infinity's $15M raise at $100M valuation signals the market is pricing this gap. The round included OpenAI and Anthropic researchers, which reads like hedging. Anthropic's $1.5B copyright settlement adds fixed costs to the closed model balance sheet. Open-source models don't carry this liability. The legal exposure alone could add $0.08 per million tokens to closed model serving costs. Apple's potential lawsuit against OpenAI's hardware plans introduces another variable. Hardware integration was supposed to be the margin recovery play—vertical stack to offset inference compression. Legal risk now threatens that timeline. The capex came home in 2023. The ROI is getting complicated.